A Practical Introduction to Monad Transformers

01 February 2015

Monad transformers combine the functionality of two monads into one. They are often used as a way to “flatten” nested monads, but are also used to enable interactions between monads that, when used seperately, would be incredibly difficult to implement.

The Task

You input a number and want to manipulate it while printing the result each time. If there were no intermediate IO operations we could use the state monad with the following state changes:

add :: Int -> State Int ()
add n = state $ \s -> ((),s + n)

subtract :: Int -> State Int ()
subtract n = state $ \s -> ((),s - n)

chain them together:

manyOperations :: State Int ()
manyOperations = do
    add 1
    subtract 3
    add 5
    add 7
    subtract 22

then get the result:

(_,result) = runState manyOperations 5 :: ((),Int)

Now let’s consider how to print the state. If we want to preserve the above chaining syntax, we need a monad where:

  1. We can do stateful computations
  2. Our IO functions have access variables inside the computation

This monad is called the state monad transformer.

The Solution

The state monad transformer is defined as:

newtype StateT s m a = StateT { runStateT :: s -> m (a,s) }

1Meaning that given an initial state s and a state transformer st, we can call runStateT st s to get a monad containing the state tuple.

The real beauty (or magic, as some would say) of this monad comes from the bind function. Let’s take a look at its definition:

(>>=) :: Monad n => StateT s n a -> (a -> StateT s n a) -> StateT s n a
m >>= k  = StateT $ \ s -> do
    ~(a, s') <- runStateT m s
    runStateT (k a) s'

Time to break it down. m is a state transformer. k is a function that takes a result of type a, and returns a state transformer. The final state transformer, when run with an initial state, does the following:

  1. Gets the result and state of running the computation in m with the initial state s
  2. Passes the result of (1) to the function k, returning a different state transformer
  3. Runs the computation created in (2) using the state returned in (1)
  4. Wraps the result

This means that we will be able to keep using a simple chained sequence of monads.

How does this relate to the problem at hand?

The monad component of the state transformer allows us to execute IO operations which have access to the state during the computation. Here is how the add and subtract functions can be written using the state transformer monad:

add :: Int -> StateT Int IO ()
add n = StateT $ \s -> do
    print (s+n)
    return ((),s+n)

subtract :: Int -> StateT Int IO ()
subtract n = StateT $ \s -> do
    print (s-n)
    return ((),s-n)

We can still chain them using the same syntax as before:

manyOperations :: StateT Int IO ()
manyOperations = do
    add 1
    subtract 3
    add 5
    add 7
    subtract 22

and run it:

main = runStateT manyOperations 5
-- output:
-- 6
-- 3
-- 8
-- 15
-- -7
  1. newtype creates a strict, isomorphic type with a single value constructor. If all that was too much, just imagine that newtype rearranges an existing type into a more pleasant one.